Curious Case of Language Generation Evaluation Metrics: A Cautionary Tale

TL;DR

This paper highlights the limitations of current automatic evaluation metrics for language generation, demonstrating their potential to misjudge system quality and urging for more careful metric selection and interpretation.

Contribution

The paper provides a comprehensive analysis of common evaluation metrics' failure modes across multiple datasets and tasks, emphasizing the need for improved evaluation practices.

Findings

Metrics often prefer system outputs over human texts

Metrics are insensitive to correct translations of rare words

Single-sentence outputs can yield high scores for entire test sets

Abstract

Automatic evaluation of language generation systems is a well-studied problem in Natural Language Processing. While novel metrics are proposed every year, a few popular metrics remain as the de facto metrics to evaluate tasks such as image captioning and machine translation, despite their known limitations. This is partly due to ease of use, and partly because researchers expect to see them and know how to interpret them. In this paper, we urge the community for more careful consideration of how they automatically evaluate their models by demonstrating important failure cases on multiple datasets, language pairs and tasks. Our experiments show that metrics (i) usually prefer system outputs to human-authored texts, (ii) can be insensitive to correct translations of rare words, (iii) can yield surprisingly high scores when given a single sentence as system output for the entire test set.